The chance of getting this right three times in a row is one chance in 8, or 12.5%. Since everyone and their dog is in the game of trying to predict presidential election success, then there must be many of them that got it right three times.

We should check whether Clinton or Trump masks have the most halloween mask sales. This metric has successfully predicted every presidential election since 1980. 6 Bizarre Factors That Predict Every Presidential Election

Which doesn’t mean that it will predict it successfully this time.

There are many other striking correlations

Age of Miss America correlates with Murders by steam, hot vapours and hot objects

More spurious correlations here: 15 Insane Things That Correlate With Each Other

In short you can’t conclude much at all from a correlation like that, especially if it is noticed after the fact.

Now, if there was a preliminary survey of all ways of predicting the US election, and there was a general consensus that this AI program was the best way to do it, then it predicted it successfully 3 times - then that would be suggestive, not very strong evidence, not nearly enough for a scientific validation, but might lead one to think that there just possibly was something worth investigating further.

But the other way around - to notice a correlation and then try to use that to say it is a likely way of predicting the election - this is not even suggestive. This sort of thing is bound to happen often just through chance.

Of course if you happen to be the developer of the AI program then it will seem striking to you. But that’s because you aren’t taking account of all the other developers of AI programs and other methods of predicting the election who failed. This is why scientists require such high standards before something is treated as a discovery in science. There are so many scientists making so many hypotheses every year that many of them are bound to make predictions that fit the data very well from time to time. The Higgs boson was only accepted as real after it was confirmed to 5 sigma meaning one chance in 3.5 million that the data was chance 5 Sigma What's That? They collect so much data for particle physics experiments that false results are common if you set the cut off point to anything much more than about 1 in 3.5 million.